Kinect-Based Substation Operation and Maintenance Action Recognition Method

Aiming at the problem of traditional substation operation and maintenance training method, this paper proposes a new method for identifying substation operation and maintenance actions based on Kinect depth camera. The method is to establish a data set through a hybrid feature extraction method of human bone joint angles and the normalized distances of joint points; establishing three models (CART, KNN and BP) to identify ten typical operation and maintenance actions of substations. The experimental results show that the accuracy of using hybrid feature recognition is higher than that of single feature, and the recognition accuracy of CART algorithm based on hybrid features reaches 98.1%. Its effect is better than KNN algorithm and BP neural network method, and it can accurately identify ten selected substation operation and maintenance actions.